Lithology Identification Of Igneous Rock Reservoirs Based On Machine Learning

The reservoir formation rules of buried hill igneous rocks in Huizhou Sag are complex,with diverse rock types and compositions.Its lithological characteristics are mostly massive structures,with variable main mineral components and altered phenomena.This leads to similar main mineral and chemical properties of lithology,similar logging response characteristics between different lithology,and great difficulty in identification.Aiming at the complex lithology and low recognition accuracy in this area.Using machine learning algorithms such as unsupervised,supervised,and semi supervised learning,three algorithms,Gaussian mixture clustering(GMM),support vector machine(SVM),and transitive support vector machine(TSVM),were selected.The main rock types,lithological characteristics,and logging response characteristics of buried hill reservoirs in Huizhou Depression are comprehensively considered.A total of6 conventional logging curve values,including natural gamma ray,density logging,compensated neutron porosity,acoustic transit time,formation true resistivity,and flushing resistivity,are selected as input data for lithologic identification of igneous rock reservoirs.Comparative analysis of the effects of three lithologic identification methods shows that GMM has the advantage of processing large amounts of data and fast speed,but the recognition accuracy is not high.Based on this,an improved GMM method based on hierarchical decomposition and principal component analysis(PCA)is proposed.Using the hierarchical decomposition idea for reference,combining the geological classification principles of igneous rocks and logging response characteristics to determine the lithology identification level.According to the hierarchical classification principle,quantitative optimization of lithologic identification sensitive parameters in each layer is used to establish a lithologic identification optimization level in the study area.After that,PCA and GMM are used to identify lithology step by step and determine its calculation function,so that a GMM model based on hierarchical decomposition and PCA improvement can be established.Its recognition accuracy is greatly improved compared to using GMM model alone,and the recognition accuracy of the model for each lithology is above 86%.Apply the improved GMM model,SVM model,and TSVM model based on hierarchical decomposition and PCA to actual wells.The results show that the accuracy of lithologic identification of the three models in actual wells has decreased by3.73%,10.66%,and 4.43% compared to the verification set,indicating that the generalization of the SVM model is insufficient.Therefore,in the early stage of lithologic identification of new wells in the research area,using an improved GMM model can ensure that good lithologic identification results can be achieved while operating at a faster speed.When there is a large amount of new well data in the later stage,using the TSVM model for lithologic identification can still maintain a good lithologic identification effect on the premise of high generalization.The research in this paper can provide a new solution to the problem of lithologic identification accuracy of igneous rock reservoirs,and provide a reference basis for accurate lithologic identification of buried hill igneous rocks in the study area.